【摘要】 长焦相机采集近距离棋盘格图像时易出现相机离焦现象，导致棋盘格图像产生散焦模糊，极大地增加了相机标定的难度，同时传统的Harris角点检测算法对散焦模糊的棋盘格图像进行角点检测的结果即使经过非极大值抑制处理也仍然存在大量冗余角点.针对上述问题，基于随机抽样一致(random sample consensus, RANSAC)算法提出一种改进的Harris-RANSAC长焦相机标定算法.首先，引入感兴趣区域将Harris角点检测的区域缩小到棋盘格区域以避免背景干扰；其次，采用随机抽样一致算法替代传统的非极大值抑制方法剔除冗余角点；最后，针对模糊棋盘格图像的特性构造新的响应函数，进行亚像素级角点定位，从而得到精确的角点坐标.结果表明，改进的Harris-RANSAC算法对模糊棋盘格图像进行角点检测时耗时短且精度较高，角点检测的反投影误差仅为0.432像素.更多还原

【Abstract】 When a telephoto camera captures close-range checkerboard images, camera defocus is easy to occur, resulting in defocusing blur in the checkerboard images, which greatly increases the difficulty of camera calibration. At the same time, the traditional Harris corner detection algorithm still has a large number of redundant corners in the corner detection of defocus blurred checkerboard images even after non-maximum suppression. To solve the above problems, an improved Harris RANSAC telephoto camera calibration algorithm based on random sampling consensus algorithm is proposed. Firstly, the region of interest is introduced to reduce the area of Harris corner detection to a checkerboard area to avoid background interference. Secondly, the random sampling consistency algorithm(RANSAC) is used to replace the traditional non maximum suppression method to remove redundant corners. Finally, a new response function is constructed based on the characteristics of fuzzy checkerboard images to perform sub pixel level corner localization and obtain accurate corner coordinates. The results show that the improved Harris RANSAC algorithm takes less time and has higher accuracy in corner detection of fuzzy checkerboard images, with a backprojection error of only 0.432 pixels for corner detection.更多还原